Robert A. Vandermeulen

I genuinely don’t understand why you think it’s bad for Dems to adapt their strategy to boost their chances of winning.

Dimension-independent rates for structured neural density estimation We show that deep neural networks achieve dimension-independent rates of convergence for learning structured densities such as those arising in image, audio, video, and text applications. More precise...

I'm happy to share our work "Dimension-independent rates for structured neural density estimation" was accepted to ICML! This furthers our new approach on sample-efficient high-dimensional nonparametric statistics!

Preprint: arxiv.org/abs/2411.15095 (old version)

w/ Wai Ming Tai and Bryon Aragam

Woop! Congrats Herr Doktor!

Major late 2000s early 2010s zombie stuff:
28 weeks later
Shaun of the dead
Zombieland
The walking dead
The last of us
Left for dead
The zombie survival guide
World war Z
Pride and Prejudice and Zombies

why were zombies so big in the 2000s?

This represents a new and theoretically grounded way of explaining how AI can perform so well on high-dimensional data. Come see our poster at NeurIPS:

neurips.cc/virtual/2024...

Excited to share 2 new papers showing how a Markov random field assumption improves density estimation—especially for dependencies in image, audio, and text data. This assumption is orthogonal to the manifold hypothesis! 📊📚

NeurIPS openreview.net/pdf?id=dWwin2uGYE
Preprint arxiv.org/abs/2411.15095